Have you tried V-Ray (RT) GPU ?

[padre.ayuso]

Yeah, I have a workstation with Dual E5-2630 v3, 64GB RAM and about 75 cores through the rendering farm.

[ramon_perez]

Hello All,

I switched completely from Adv. to RTGPU (single dual-xeon mobo / 4x1080ti vs. previous 3x dual xeon ws/mini farm)

Main reasons: cpu heat generation, power consumption, huge pc cases with only 2cpus each, reducing render times... ALL traded over image quality.

---------------------------------
CONS:

a LOT of tweaking of shaders for images to look less "plasticky" than when rendered in Adv.

The render engine can be unreliable on heavy scenes, so one needs to be very careful with meshes and do a lot of optimization. So you can spend troubleshooting and recovering a lot of time...

The VrayWireColor render element is not supported in RT! which makes it painful to work with object selections in Photoshop when using alternatives as RenderID.

Vray Light "Portal" can generate (squared looking) artifacts closer to the camera when using LC + BF. It doesn't generate the artifacts On BF+BF, but it still can generate less noticeable (tiny) artifacts.

Option "Override material" in the Global Switches Tab when using IPR in RTGPU is not supported.

VrayOverrideMtl is not supported on RT GPU when using IPR, a material that could be essential in some cases.

IPR doesn't use LC+BF so it can be misleading on light balancing when using LC+BF for a final render.

PRO:

Render speeds are A LOT faster overall, RTGPU takes a TON of time to load textures and compile geometry, BUT, in my experience, a heavy scene (with tons of trees and geometry) when optimized properly, can render 4K or larger images on a single computer (using 4x1080ti) in less time than it takes 3 dual-xeon pcs (10 -12cores) to calculate GI only. My 2 cents.

--------------

If anyone here could shed some light on any of the features that are not supported yet, it would be great.

Peace!

[Alex_M]

If anyone here could shed some light on any of the features that are not supported yet, it would be great.

You can check this in the help files.

[Vladislav Vulchev]

Hi ramon_perez,

VrayOverrideMtl and Override material Global Switch should work fine. Do you have some specific scenario in which they don't?

[ramon_perez]

Hello Vladislav,

While doing interactive rendering using the IPR on RT-GPU, If I try to use Override Material in the Global Switch with a gray standard material, nothing happens in the scene. All materials remain unchanged. However if I choose to do a final rendering of the scene "F9" then it works, but I Loose the interactivity of IPR. (Note:It works on active shade)

Same goes for VrayOverrideMtl. I can double check with F9 if it works, however I was trying it on IPR for GI override (Gray standard over a dark wood floor) and it did not work. Thx!

------------------------

Alex_M will check out the link. Thanks!

[danio]

CONS:

a LOT of tweaking of shaders for images to look less "plasticky" than when rendered in Adv.

I wonder if this is due to Glossy Fresnel not being support by RT GPU yet? Or is it something else about the way RT works vs Adv?

[thanulee]

Sorry i have to ask this again, whats the difference with active shade and IPR in RT? isnt it the exact same thing? ty

[matanov]

While doing interactive rendering using the IPR on RT-GPU, If I try to use Override Material in the Global Switch with a gray standard material, nothing happens in the scene. All materials remain unchanged. However if I choose to do a final rendering of the scene "F9" then it works, but I Loose the interactivity of IPR. (Note:It works on active shade)

There are separate override materials for Production and ActiveShade. Hitting F9 will launch the Production render and it will use the override from the Global Switches, while using the IPR button (if RT is set as Production) will launch ActiveShade which will use its own override mtl set in the Render setup of ActiveShade.

Same goes for VrayOverrideMtl. I can double check with F9 if it works, however I was trying it on IPR for GI override (Gray standard over a dark wood floor) and it did not work. Thx!

Can you check that again? It should work just fine, if not please send me the scene.

[matanov]

Sorry i have to ask this again, whats the difference with active shade and IPR in RT? isnt it the exact same thing? ty

https://docs.chaosgroup.com/display/...IPR+in+3ds+Max

https://forums.chaosgroup.com/forum/...102#post841102

[ramon_perez]

OK Thanks for the clarification. I didn't know IPR was controlled under the active shade settings. It all makes more sense now.

I have checked again and came to conclusion that VrayOverrideMtl wasn't working because I was using it over a VrayBlend Material as a Base. Once I changed the VraybBlend to a Normal material, the GI override worked.

------------------------

is VrayWireColor render element going to be supported in the future?



Thx!

[Vladislav Vulchev]

VrayOverrideMtl wasn't working because I was using it over a VrayBlend Material as a Base

We are aware of this problem. For now using VrayOverrideMtl in VrayBlend Material or vise versa doesn't work.

[padre.ayuso]

Hello All,

Render speeds are A LOT faster overall, RTGPU takes a TON of time to load textures and compile geometry, BUT, in my experience, a heavy scene (with tons of trees and geometry) when optimized properly, can render 4K or larger images on a single computer (using 4x1080ti) in less time than it takes 3 dual-xeon pcs (10 -12cores) to calculate GI only. My 2 cents.

Thanks Ramon! That's good data. Just to clarify, 3 dual xeon pcs (10-12 cores) per CPU or per PC? When you said to calculate GI only, do you have some comparison for a full scene, from beginning to end? It used to be that the GI was calculated from the CPU before it went through to GPU, perhaps this has changed over the last few months, don't know, have not been using RT too much hence the data would not be very relevant to how fast RT can handle vs. CPU. I'm still interested to do my work on GPU and I may be investing in render farm in the very near future, so the decision to go RT or stay Adv. is coming up soon.

Also, anyone knows where to get GPU clusters/render farms? If going RT, I'm not interested in having huge workstations with 4 cards, I'm having having 2U machines in my server room with 4 cards in them.

Thanks for any data on this.

[ramon_perez]

Hello padre.ayuso

3x dual xeons with 10 12 cores PER cpu, these were V2 xeons. I have gotten rid of two machines already. but I remember rendering times of easily 5 hours using all 3 (distributed) using IRMAP+LC. Then, using only 2x 1080ti, I re-rendered the image after some minor tweaks requested by client. It took almost 3 hours to finish on those 2x 1080ti using LC+BF. Now, using four cards you can do the math...

Recently I've been rendering using BF+BF and I see 99% usage on four cards and only ~15% usage on CPU when rendering, so I'm 'assuming' all GI calculations are being done on the GPUs. (anyone please clarify)

I would say that in my experience, you start noticing significant speed advantages over multiple cpus when using 3 high end cards or more. I think on a gpu farm the difference would be huge.

[tue_nguyen]

I not sure but when i test with VrayRT, the lighting is softer than Vray advance. I'm feeling it's similar corona lighting. It doesn't support some material example FastSSS. But the most important is we don't have wirecolor channel! I don't know how to post product without it.
I tested with CPU RT. With GPU RT let's try with Vray proxy bitmap, it's realy awesome for save your vram (of gpu).

[padre.ayuso]

Hello padre.ayuso

3x dual xeons with 10 12 cores PER cpu, these were V2 xeons. I have gotten rid of two machines already. but I remember rendering times of easily 5 hours using all 3 (distributed) using IRMAP+LC. Then, using only 2x 1080ti, I re-rendered the image after some minor tweaks requested by client. It took almost 3 hours to finish on those 2x 1080ti using LC+BF. Now, using four cards you can do the math...

Recently I've been rendering using BF+BF and I see 99% usage on four cards and only ~15% usage on CPU when rendering, so I'm 'assuming' all GI calculations are being done on the GPUs. (anyone please clarify)

I would say that in my experience, you start noticing significant speed advantages over multiple cpus when using 3 high end cards or more. I think on a gpu farm the difference would be huge.

That's great! Except I have read in some other thread that the speed is not linear and that going above 4 cards will not give you much of a speed improvement. I'm sure Vlado, Blago and team are working on getting this sorted.